The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from external project files and a memory retrieval tool to dictate agent behavior and communication style.
Ingestion points: The skill reads from USER.md (persona/preferences), CURRENT.md (active session constraints), and a memory database via the get_context tool or memory context CLI command.
Boundary markers: The instructions do not define specific boundary markers or delimiters for the ingested content, relying on simple markdown headers for separation.
Capability inventory: The context retrieved is intended to influence high-privilege agent actions including code mutation, architectural planning, and refactoring.
Sanitization: There is no evidence of sanitization or filtering for instructions embedded within the source files. The instruction to "Strictly mirror the persona defined in USER.md" provides a direct path for data-embedded instructions to override agent behavior.
[COMMAND_EXECUTION]: The skill provides CLI fallback commands for retrieving context when MCP tools are unavailable.
Evidence: The skill includes commands such as memory context, memory current get, and memory user get for local execution.

muse-ground